Perks and Quirks: Using the NCDB Participant User File (PUF) for Outcomes Research Daniel J. Boffa, MD Thoracic Surgery Yale University School of Medicine
None Disclosures
Why Me? No formal affiliation with NCDB
Why Me? No formal affiliation with NCDB Outcomes research with NCDB PUF x 4 years
Why Me? No formal affiliation with NCDB Outcomes research with NCDB PUF x 4 years MANY Manuscript Rejections Good sense of perceptions Made lots of mistakes
NCDB Participant User File (PUF) American College Surgeons American Cancer Society
NCDB Participant User File (PUF) De-identified, comprehensive data set Site Specific (colon, breast, prostate) Patient care research Clinician-investigators at CoC accredited cancer centers
Who is in NCDB Hospital with CoC Accredited Cancer Center 1,500 (30% of U.S. Hospitals) Report All Newly diagnosed CA (Obligatory)
Who is in NCDB Hospital with CoC Accredited Cancer Center 1,500 (30% of U.S. Hospitals) Report All Newly diagnosed CA (Obligatory) Treatment or Diagnostic Evaluation (i.e. not just pathology review)
Who is in NCDB Hospital with CoC Accredited Cancer Center 1,500 (30% of U.S. Hospitals) Report All Newly diagnosed CA (Obligatory) Treatment or Diagnostic Evaluation (i.e. not just pathology review) 70% of CA patients in U.S.
Who is NOT in NCDB PUF Veterans from VA hospitals
Who is NOT in NCDB PUF Veterans from VA hospitals A few Top Rated cancer centers
What DATA captured Sociodemographics Cancer Treatment Survival
What DATA captured Sociodemographics Cancer Treatment Survival First course of treatment
Age Race Sex Sociodemographics Included Insurance Regional Statistics Distance traveled Facility type (i.e. teaching)
Age Race Sex Sociodemographics Included Insurance Regional Statistics Distance traveled Facility type (i.e. teaching)
Sociodemographics Included Age Race Sex Insurance Regional Statistics Regional Distance traveled Statistics Distance Facility type traveled (i.e. teaching) Not Weight (BMI) Smoking Performance Status Facility characteristics (# beds, specialty services)
Included Modified Charlson Comorbidity Index 0, 1, >2 Health
Health Included Modified Charlson Comorbidity Index 0, 1, >2 Not Specific conditions (emphysema) Severity (use of inhalers vs. Oxygen) How established (CT scan or spirometry)
Example SBRT vs. Lobectomy NCDB PUF Stage I Lung cancer patients Treatment Stereotactic Body Radiation Therapy (SBRT) Surgery (lobectomy, wedge, segment)
Example SBRT vs. Lobectomy NCDB PUF Stage I Lung cancer patients Treatment Stereotactic Body Radiation Therapy (SBRT) Surgery (lobectomy, wedge, segment) Propensity Matched Cohorts
Figure 2 Surgery SBRT Journal of Thoracic Oncology 2015 10, 1776-1784DOI: (10.1097/JTO.0000000000000680) Copyright 2015 International Association for the Study of Lung Cancer Terms and Conditions
Figure 2 Surgery Surgery Better than SBRT SBRT Journal of Thoracic Oncology 2015 10, 1776-1784DOI: (10.1097/JTO.0000000000000680) Copyright 2015 International Association for the Study of Lung Cancer Terms and Conditions
Figure 2 Surgery SBRT < 30% 5y Survival Journal of Thoracic Oncology 2015 10, 1776-1784DOI: (10.1097/JTO.0000000000000680) Copyright 2015 International Association for the Study of Lung Cancer Terms and Conditions
Figure 2 Surgery SBRT < 30% 5y Survival MUCH lower than published case series and trials Journal of Thoracic Oncology 2015 10, 1776-1784DOI: (10.1097/JTO.0000000000000680) Copyright 2015 International Association for the Study of Lung Cancer Terms and Conditions
Something not right
Adjustment Strategies Cox proportional hazards
Only can adjust for variability that is captured by data collected
Comorbidity Number not severity
Emphysema
Emphysema
Performance Status Kendall Simmons NFL Guard Diabetes
Performance Status
Only can adjust for variability that is captured by data collected Particularly Poor Health Not able to be identified/stratified
Adjustment Strategies Cox proportional hazards Propensity Matching
Adjustment Strategies Cox proportional hazards Propensity Matching
SBRT practice Not healthy enough for surgery Poor health = competing survival risk
No Match!
NCDB Perks Reason no surgery field Not part of initial treatment plan
NCDB Perks Reason no surgery field Not part of initial treatment plan Surgery not recommended because of poor health
NCDB Perks Reason no surgery field Not part of initial treatment plan Surgery not recommended because of poor health Exclude particularly unhealthy patients from nonsurgical cohort
NCDB Perks Reason no surgery field Not part of initial treatment plan Surgery not recommended because of poor health Surgery was recommended, but patient refused.
NCDB Perks Reason no surgery field Not part of initial treatment plan Surgery not recommended because of poor health Surgery was recommended, but patient refused. Study subset of nonsurgical cohort healthy enough for surgery
Yale Study in NCDB SBRT vs Lobectomy
Yale Study in NCDB SBRT vs Lobectomy Healthy Patients All comorbidity free Excluded SBRT patients who not healthy enough for surgery ( Reason no surgery field)
Yale Study in NCDB SBRT vs Lobectomy Healthy Patients All comorbidity free Excluded SBRT patients who not healthy enough for surgery ( Reason no surgery field) Propensity matched 1200 pairs
Matched Healthy Stage I NSCLC Survival SBRT Lobectomy Months Rosen et al, J Thor Cardiovasc Surg (in print)
Yale Study in NCDB SBRT vs Lobectomy Healthy Patients All comorbidity free Excluded SBRT patients who not healthy enough for surgery Only SBRT Patients that refused recommended surgery
SBRT Refused Recommended Sx Lobectomy Survival SBRT Months Rosen et al, J Thor Cardiovasc Surg (in print)
SBRT Refused Recommended Sx Lobectomy Survival SBRT 5y Survival 29% 40% = Health Related Bias Months Rosen et al, J Thor Cardiovasc Surg (in print)
Included Histology Clinical stage Pathologic Stage # of nodes Margin status Site specific factors (e.g. CEA for colon) Cancer
Included Histology Clinical stage Pathologic Stage # of nodes Margin status Site specific factors (e.g. CEA for colon) Cancer
Cancer Included Histology Clinical stage Pathologic Stage # of nodes Margin status Site specific factors (e.g. CEA for colon) Not Staging tests (CT, PET) Tumor board consensus stage Mutations (EGFR lung cancer)
Included Surgery Detailed Chemotherapy Multi-agent vs single Radiation Highly detailed Palliative Sequence Pre/post operative Treatment
Treatment Surgery Included Detailed Chemotherapy Multiagent v single Radiation Highly detailed Palliative Not Chemotherapy type (platinum) Concurrent versus sequential chemoradiation
NCDB Quirk - Treatment Every CoC Hospital reports Every patient
NCDB Quirk - Treatment Every CoC Hospital reports Every patient Reporting hospital must include care outside their hospital
NCDB Quirk - Treatment Every CoC Hospital reports Every patient Reporting hospital must include care outside their hospital Duplicate reporting NCDB algorithm to identify
NCDB Quirk - Treatment Every CoC Hospital reports Every patient Reporting hospital must include care outside their hospital Duplicate reporting NCDB algorithm to identify NCDB takes most complete report (does not compile a composite)
Included 30-day surgical mortality 90-day surgical mortality Unplanned readmissions Adverse Events
Included 30-day surgical mortality 90-day surgical mortality Unplanned readmissions Adverse Events
NCDB Quirk - Readmissions Only readmissions to reporting hospital
Readmissions after colorectal surgery
Readmissions after colorectal surgery Relatively uniform around facility
Readmissions after colorectal surgery Decreased significantly If patients Traveled >160 miles
Adverse Events Included 30 day mortality surgery Unplanned readmissions (to reporting hospital) Not Specific surgical complications Adverse events with nonsurgical Tx
Included All-Cause Survival Follow up
Follow up Included All-Cause Survival Not Recurrence-Free Survival Cause of death Cancer-specific survival Quality of Life
Transitions over time 1988 2017
Transitions over time Added Charlson Co-morbidity Required reporting of radiation Tx 1988 2017 2003
Transitions over time Registrars code Clinical Stage from chart review (dramatically decreased missing cases) 1988 2017 2008
Transitions over time Surgical approach (minimally invasive, robot) 1988 2017 2010
Transitions over time Shortened embargo on survival data 7y 3y (PUF released in 2015 has survival through 2012 cases) 1988 2017 2011
Transitions over time Field to indicate if treatment occurred at reporting facility (previously, annual volume estimates were ambiguous) 1988 2017 (retro) 2014
Other NCDB Quirks Age (younger more likely captured) 73% of pts <65 captured 63% of pts >65 captured
Other NCDB Quirks Age (younger more likely captured) 73% of pts <65 captured 63% of pts >65 captured Racial disparities (Hispanic less likely) 50% of cases in pts of hispanic origin captured 65% of cases (white, black)
Other NCDB Quirks Age (younger more likely captured) 73% of pts <65 captured 63% of pts >65 captured Racial disparities 50% of cases in pts of hispanic origin captured 65% of cases (white, black) Regional variation 89% of Delaware 27% of Arizona
NCDB Quirk Clinical Impressions Clinical Stage Reason no surgery field Surgery recommended Surgery not recommend
NCDB Quirks Clinical Impressions
NCDB Quirks Clinical Impressions
Other NCDB Perks >90% follow up User Friendly Accessible NCDB consultants Free/already paying for it
Conclusions NCDB PUF has enormous potential Transparent about limitations Changes over time critical to recognize report version of PUF
http://ncdbpuf.facs.org/node/25 9 Data dictionary